Applied legal AI · course approach

AI legal engineer

Build an AI-assisted legal workflow you can explain, test and defend. Start with the client’s problem, compare what the models produce, and take responsibility for the advice.

The working method

Design the work. Check the result.

  1. Frame the taskTurn an open-ended request into an explicit question, with a client, an audience and a decision in view.
  2. Choose the evidenceRead the document alongside the client file. Decide which facts the model needs, what is missing and what should not be disclosed.
  3. Compare modelsRun the task with different models. Identify disagreements and shared assumptions, then explain how you judged the answers.
  4. Verify independentlyCheck sources, unsupported claims and omissions. Separate generating an answer from checking it, and record the work in a prompt log.
  5. Defend the advicePresent a reasoned recommendation, respond to challenge and explain what a lawyer must check before putting their name to it.

Erasmus School of Law · Rotterdam

A legal problem with AI in the workflow

JJ’s teaching for Erasmus School of Law master’s students used group work and a financing case to test how AI findings are extracted, compared and challenged.

The assignment required each student to work with a model and another group member to verify using a different model. Teams paired a term sheet with a client file, documented their prompts, investigated unreliable claims and omissions, and presented advice for challenge.

The quality of the advice depends on the evidence you supply, the questions you ask and the checks you perform.

Delivery

Adapted to the room

The method can be taught to students and working lawyers. Professional sessions are available by arrangement for firms, in-house legal teams and professional bodies. Scope, duration and exercises are agreed for each delivery.

Enquire about a session

Cut the Crap — the briefing

The developments that matter, tested against the documents.

Subscribe

Delivered via Substack. No spam — unsubscribe anytime.